docs(agents): adopt the generated git-workflow block #940

Merged

coilyco-ops merged 3 commits from aos/claude/sy55 into main

2026-08-26 02:38:22 +00:00

coilyco-ops commented

2026-08-26 02:36:07 +00:00

Owner

Closes #937 for this repo.

AGENTS.md carried a hand-written stamp saying the merge on pull-request-and-merge stays director-gated. The canonical renderer in agentic-os says the opposite and names that exact reading as the inversion it was written to correct: the agent that authored the code merges its own pull request, and reading the lane the other way leaves finished work sitting unmerged. The stamp also claimed byte-identical parity across the five PR-lane repos, which was false, because agentic-os carries the generated block instead.

Rendered by agentic-os/scripts/apply-git-workflow.py from the lane this file already declares in its own frontmatter, so nothing per-repo chose it. The applier also strips the legacy one-line stamp on its own.

The cap raise

The block costs about 2.3 KB and put AGENTS.md at 8133 chars against an 8000 cap. The cap opts up to 8500 for the same reason it already opted up once for the engineer/QA prohibition: a consumer repo cannot split, trim, or reword machine-generated text, so the block either fits or the hook blocks a file this repo does not author. The comment says to raise it only for another block in that category.

What this does not do

The drift is fleet-wide, not local. A dry run of the applier across the 20 checked-out repos reports agentic-os as the only one already current and 18 that would change, on both lanes. The reason is structural: agentic-os wires the git-workflow drift gate as a local hook, and the managed hook set that apply-agentic-os-hooks.py writes into consumer repos does not include it, so no other repo has ever been gated against this drift.

Closing that has an ordering constraint worth stating before anyone starts. Adding git-workflow to the fleet hook set first turns 18 repos red until each one takes its block, so the blocks land first and the gate follows. Each repo also pays the same ~2.3 KB against its own AGENTS.md cap, so some will need the same raise this pull request makes here.

That rollout belongs in this repo as ansible, not as a hand run of the authored applier, per the authoring-vs-rollout split the applier docstring names. Left on the issue rather than done here.

Co-Authored-By: Claude Opus 5 (1M context) noreply@anthropic.com

Closes https://forgejo.coilysiren.me/coilyco-flight-deck/infrastructure/issues/937 for this repo. `AGENTS.md` carried a hand-written stamp saying the merge on `pull-request-and-merge` stays director-gated. The canonical renderer in agentic-os says the opposite and names that exact reading as the inversion it was written to correct: the agent that authored the code merges its own pull request, and reading the lane the other way leaves finished work sitting unmerged. The stamp also claimed byte-identical parity across the five PR-lane repos, which was false, because agentic-os carries the generated block instead. Rendered by `agentic-os/scripts/apply-git-workflow.py` from the lane this file already declares in its own frontmatter, so nothing per-repo chose it. The applier also strips the legacy one-line stamp on its own. ## The cap raise The block costs about 2.3 KB and put `AGENTS.md` at 8133 chars against an 8000 cap. The cap opts up to 8500 for the same reason it already opted up once for the engineer/QA prohibition: a consumer repo cannot split, trim, or reword machine-generated text, so the block either fits or the hook blocks a file this repo does not author. The comment says to raise it only for another block in that category. ## What this does not do The drift is fleet-wide, not local. A dry run of the applier across the 20 checked-out repos reports agentic-os as the only one already current and 18 that would change, on both lanes. The reason is structural: agentic-os wires the `git-workflow` drift gate as a local hook, and the managed hook set that `apply-agentic-os-hooks.py` writes into consumer repos does not include it, so no other repo has ever been gated against this drift. Closing that has an ordering constraint worth stating before anyone starts. Adding `git-workflow` to the fleet hook set first turns 18 repos red until each one takes its block, so the blocks land first and the gate follows. Each repo also pays the same ~2.3 KB against its own AGENTS.md cap, so some will need the same raise this pull request makes here. That rollout belongs in this repo as ansible, not as a hand run of the authored applier, per the authoring-vs-rollout split the applier docstring names. Left on the issue rather than done here. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>

coilyco-ops added 2 commits

2026-08-26 02:36:07 +00:00

fix(claude): drop the Write memory deny the harness cannot match

CI / lint (pull_request) Successful in 42s

Details

TruffleHog / Scan for secrets (pull_request) Successful in 50s

Details

57b6356e5a

Claude Code matches file permission rules on the Edit(path) spelling, which
covers Write, Edit, MultiEdit, and NotebookEdit alike, so this repo-local copy
carried a dead rule the harness warned about at session start. The sibling
Edit rule already denies the whole memory surface, so enforcement is unchanged.

The fleet writer this mirrors loses the same rule in agentic-os, which also
gains a prune pass so hosts that already converged it get it removed.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Co-authored-by: Kai Siren <coilysiren@gmail.com>
Co-authored-by: Claude <noreply@anthropic.com>
Agent-Role: platform

docs(agents): adopt the generated git-workflow block

TruffleHog / Scan for secrets (pull_request) Successful in 5s

Details

CI / lint (pull_request) Successful in 41s

Details

ed2963baa8

AGENTS.md carried a hand-written one-line stamp saying the merge on
pull-request-and-merge stays director-gated. The canonical renderer in
agentic-os says the opposite, and names that exact reading as the inversion it
was written to correct: the agent that authored the code merges its own pull
request, and reading the lane the other way leaves finished work unmerged.

The stamp also claimed to be byte-identical across the five PR-lane repos. It
was not, because agentic-os carries the generated block instead.

Rendered by agentic-os/scripts/apply-git-workflow.py from the lane this file
already declares in its own frontmatter, so no per-repo argument chose it. The
block costs ~2.3 KB, which puts the file past its char cap, so the cap opts up
for the same reason it already opted up once: a consumer repo cannot trim text
it does not author.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Co-authored-by: Kai Siren <coilysiren@gmail.com>
Co-authored-by: Claude <noreply@anthropic.com>
Agent-Role: platform

coilyco-ops referenced this pull request

2026-08-26 02:36:27 +00:00

AGENTS.md tells agents the merge is director-gated, which inverts the pull-request-and-merge lane and strands finished work #937

coilyco-ops added 1 commit

2026-08-26 02:37:23 +00:00

Merge branch 'main' into aos/claude/sy55

TruffleHog / Scan for secrets (pull_request) Successful in 4s

Details

CI / lint (pull_request) Successful in 41s

Details

c66b573e38

coilyco-ops merged commit fae5042296 into main